BREAKING
Kimi K3 Ties Claude and GPT on Agent Arena
0
Agent Arena rank
0
pts
Frontend Code Arena #1
0
places gained vs K2.6
0
T
total parameters
0
M
token context
Uneven Agentic Profile
Task Success
1
Praise/Complaint
3
Steerability
14
Bash Recovery
17
Strengths and Gaps
Strengths
●
Frontend coding leader
●
Pricing 1/3 to 1/5 of rivals
●
Whole-repo refactoring
Gaps
●
Trails Claude on judgment
●
More hallucinations
●
Weak error recovery
Full Weights Due July 27
AI NEWS BLITZ
Moonshot AI's Kimi K3 has surged onto the Agent Arena leaderboard, matching the top closed models.